Comprehensive protocols for incident identification, forensic evidence verification, and disputed causation management.
The AIDEATHS Database is an independent, non-commercial public research repository. An incident is documented in our database only when it meets three strict criteria:
The incident resulted directly in human loss of life or catastrophic, permanent physical or psychological injury requiring emergency medical care.
An autonomous algorithm, machine learning model, companion chatbot, or automated physical system played an active role in the failure sequence.
The incident is supported by documented regulatory records (such as NTSB or NHTSA), coroner inquest findings, sworn court testimony, or published investigative reporting.
Every substantive claim in an incident dossier is tied to primary sources. We categorize and weigh all supporting evidence according to four rigorous evidentiary tiers:
Formal reports and technical accident reconstructions published by statutory authorities, including the NTSB, NHTSA, OSHA, FAA, UK MHRA, EU AI Office, and Japan NIOSH.
Sworn depositions, official coroner verdicts, formal court rulings, grand jury filings, and certified forensic toxicology and autopsy documentation.
Studies published in recognized academic journals (IEEE, ACM, JAMA) or in-depth investigations conducted by reputable media organizations with named sources.
Post-mortem disclosures, CAN bus telemetry, and sensor records released directly by the artificial intelligence model developers or vehicle manufacturers.
Establishing legal and technical causation in complex autonomous systems can be disputed. When an AI system's direct causal link is formally contested by manufacturers, defense counsel, or independent forensic analysts:
We attach a prominent "Causation disputed" badge to the dossier and outline the opposing technical arguments until a final regulatory or judicial ruling is issued.
To request a dossier revision or submit technical corrections, authorized representatives may contact the editorial team with supporting docket citations.
If you possess primary documentation regarding an unlisted AI-related fatality or severe incident, you may submit it to our editorial desk for verification.
• Incident timestamp & precise geographic coordinates
• Identified AI system, model architecture, or hardware platform
• Minimum of one verifiable Tier 1–3 primary source citation
• Summary of human casualties and injury severity ratings
All submissions are reviewed by our editors against primary sources before publication. Whistleblower identities and communications remain strictly confidential under cryptographic security standards.
+ Open Incident Submission Form